Research

Make weak ideas fail early.

Research quality is governed by how ideas are registered, challenged, validated, preserved and admitted—not by how attractive the final chart appears.

01 · Discipline

Every material result should have a recoverable lineage from premise to data, code, experiment, decision and review.

Controlled experimentation.

Hypothesis registration

The economic or behavioural premise is stated before implementation is tuned. Search scope and material assumptions should be declared rather than reconstructed later.

Data and leakage controls

Feature availability, dataset provenance and look-ahead risk are treated as first-class research controls. A useful model built on contaminated information is still invalid evidence.

Out-of-sample discipline

Validation boundaries exist to constrain researcher discretion. Out-of-sample evidence is not used to rescue or retune a failing candidate.

Negative research memory

Rejected hypotheses and failed variants remain part of the research record. This reduces repeated dead ends and prevents survivorship from masquerading as process quality.

Admission, not celebration

A candidate that survives one gate advances to the next applicable gate. No isolated result is treated as automatic production authority.

02 · Public boundary

Internal evidence stays internal.

This website describes the philosophy and control architecture. It does not publish live signals, positions, proprietary strategy specifications, internal OOS evidence or research-system credentials.